Development of a Body-Worn Textile-Based Strain Sensor: Application to Diabetic Foot Assessment

1. Introduction

Globally, 18.6 million people are affected annually by Diabetic Foot Ulcers (DFUs) [1]. In the UK, DFUs affect up to 25% of people with diabetes in their lifetime, with less than 50% of patients alive and ulcer-free 12 months after a DFU diagnosis [2] and almost 60% of patients experiencing recurrence within three years [3,4]. The integration of sensors within the human-interface boundary creates a paradox known as the observer effect [5]. By attempting to measure an environment by introducing a sensor, the environment under ivestigation is altered [6]. In wearables, a synergy exists between environment, behaviour and the measurement of interest. In the case of physical human interaction, this can potentially significantly impact the response observed. In the specific case of wearables for assessing the interface between foot and shoe, occupying additional internal shoe volume may reduce the space naturally available increasing the pressure observed. This patient group is particularly susceptible to changes in the shoe environment. Within the literature, wearable sensing technologies have been implemented that could be used to monitor DFU-forming conditions. Including monitoring temperature [7], tribo-electric nano-generators [8], perspiration [9] and electrocardiogram [10] among others. However, the primary focus is on interaction forces due to their influence on (DFU) formation. Diabetes can cause tissue structural changes [11], including increased sensitivity to applied forces, resulting in tissue degradation and DFU formation [11,12]. Prolonged exposure to forces normal to the skin has been linked to the formation of pressure ulcers [13,14,15]. Over the last 20 years, work has identified a coupling between normal and shear forces in DFU formation [16,17]. Additionally, nerve damage can result in diabetic peripheral neuropathy, causing reduced patient pain sensitivity and, therefore, the ability to feel and react to an applied force [18]. This set of circumstances means sensors play a critical role in DFU prevention, demonstrated by the history of foot-based sensing systems aimed at DFU prevention.

Insoles have been implemented in assessing foot performance since the early 1990s with systems such as the Tekscan F-SCAN [MA, USA] and Novel Pedar [Munich, Germany] [19,20]. Since then, a range of force-sensitive resistor-based approaches have been produced [21,22,23]. Key issues affecting force-sensitive resistors include drift and part-to-part repeatability. Alternate pressure mapping methods developed include multimodel arrays [24], alternative piezoelectric materials [25], capacitive sensors [26] and fibre optic systems [27,28]. These systems measure the vertical pressure acting perpendicular or normally to the foot’s surface, negating the shear stress components acting parallel to the skin’s surface.

Within the literature, researchers have reported the development of insoles capable of detecting shear since the early 1980s [29,30,31]. In recent years, a range of technologies have been implemented within insoles to measure shear, such as inductive sensor arrays [32,33], magnetic [34,35], capacitive [36], and microelectromechanical [37]. During gait, up to $1.5\times$ bodyweight is placed through the foot, resulting in a challenging environment to introduce sensitive electronics. A proposed solution is the STAMPS insoles, where a cumulative measure is taken to help identify strain-prone plantar regions [38]. While some systems, such as STAMPS, focus on laboratory or clinic-based measurement for diagnosis or ongoing clinical engagement to identify at-risk areas, they do not provide real-time feedback. This paper therefore, focuses on real-time shear force measurement in daily life. Information would be fed back the wearer so that they can modify foot loading that would otherwise lead to DFU development. The form and method of feedback will not be investigated at this stage.

Not limited to diabetic patient measurement, in recent years, a range of methods have been developed for investigating shear, such as magnetic [39], inductive [33,40] sensors to measure plantar shear [41]. So far, the critical drawback of these methods has been the form factor and the ability to place said sensors within the human interface region. A more in-depth review of sensing technologies, including preliminary replication studies, is presented in Section 2.2, which assesses methods with respect to this application.

A fundamental limitation of insoles is they are limited to use within a shoe. Although it may not be clinically recommended, users often remove their shoes in the comfort of their homes [42]. A textile-based approach with sock-integrated sensors enables continuous tracking with users wearing the sensing technology as they wish. Additionally, insoles are not compatible with every shoe style due to the available volume within the shoe. Although some insoles are thin, they will occupy additional volume, potentially increasing pressure on the bridge of the foot. The increased tissue sensitivity in people with diabetes increases the importance of technologies within the human interface. Using a textile-based approach allows us to swap items already present in everyday life, avoiding the additional material and minimising the interference caused by the technology within the human interface region. In some instances, people with diabetes may be required to wear custom orthotic insoles and can not use an instrumented insole. The textile-based approach, therefore, increases the number of people who can benefit from monitoring. While a sensorised shoe could be implemented, significant design and performance constraints and requirements exist. Patient engagement with technology will determine how likely they are to use it daily. A shoe-based system would be limited in design, reducing the individual’s ability to choose a style, a major concern for patients [42].

One of the critical challenges faced in this research field is the direct measurement of shear, which typically requires the parallel movement of two planes [40,43,44]. This leads to inherent design challenges regarding system miniaturisation and robustness. Instead, the measurement of strain is proposed as a proxy for shear. In the case of measuring shear applied to a foot, it is assumed that shear leading to tissue deformation will simultaneously generate a strain response within a textile in contact with said soft tissue. Therefore, infering a shear force has been applied. This assumption was the primary mechanism of action taken forward for selecting and evaluating sensing modalities and materials in this research.

In this paper, a textile-based strain sensor in proposed where strain is taken as a proxy of shear at the skin in wearable devices. Section 2, sets out the specification of a strain sensor for use within the physical human interface to assess DFU risk. Currently available techniques are then reviewed with some replications before selecting the material taken forward in sensor development. The evaluation process included some preliminary testing and was specific to the application of this study; therefore, it was intentionally separated from the literature review. Section 3, sets out sensor development, including the manufacturing techniques and parametric assessment, to identify a design best suited for the application. Section 3.4, evaluates sensor performance and demonstrate repeatability both in sensor output and manufacture. Finally, Section 4, applies the sensor to a knitted substrate, consistent with the application, and demonstrate a sensor proof of concept.

2. Materials and Methods

2.1. Scope

This research focuses on measuring shear forces within a smart sock to inform interventions preventing DFU formation. Our approach is focused on the development of a system capable of integration within the existing industrial knitting machines used to fabricate socks at scale for the mass market. This focus led us to develop an application-specific scope for assessing the potential technologies available to solve this problem. The specification summarises the requirements (Table 1).

People with diabetes have an increased risk of developing peripheral arterial disease, which can inhibit blood flow and oxygen supply to the extremities [3]. Hyperglycemia can lead to the degradation of soft tissues in the extremities, leaving them more vulnerable to injury [11]. Where nerve endings are damaged, neuropathy can develop with reduced sensation [18]. Combined with the proximity of the sensing technology to the skin and any potential openings, the method’s biocompatibility is essential. While sealing from the environment can be undertaken, biocompatibility at failure must be considered to ensure no harmful materials are released where they could affect biological systems. Due to tissue sensitivity experienced by people with diabetes, even seams in socks can lead to DFU formation [42]. Thus, care must be taken in both material choices and ensuring the designs do not produce artefacts such as ridges within the sock material.

Although starting at a conceptual phase, the project focused on methods ready for translation and scale-up to higher Technology Readiness Levels (TRLs) without significant additional development. Accordingly, the approach focused on sensors fabricated from commercially available materials, as opposed to materials development, which brings potential time, cost and regulatory constraints. The overall technology should also be scalable and low-cost per unit to facilitate adoption within sock garments.

Through clinical and patient stakeholder engagement, the wider project identified what a DFU prevention system should include [42]. Combined with the above points, these findings informed the specification (Table 1).

2.2. Technology Evaluation

During the initial investigations, the authors set out to implement or adapt pre-existing strain sensor methods published in the scientific literature [31,45], reviewing them against the specification (Table 1) for appropriateness. Various technologies and approaches were investigated, covering broader electrical interactions such as inductive, capacitive and resistive, followed by narrower material selection investigations such as carbon and silver-based approaches. Although optical fibre-based sensing in an established sensing method [46] it was excluded due to the fragility of fibres, plantar-foots high force environment (SID 1) and the lack of reinforcement in the textile substrate, marking optical fibres as inappropriate for the application. Magnetic approaches [47,48] were excluded due to the requirement to include hall-effect chips creating pressure points within the textile (SID 7). A graphical summary showing the sensor materials and methods that were explored is shown in Figure 1.

2.2.1. Self-Generating

Self-generating sensor methods focused on triboelectric nano-generators were evaluated. A fundamental mode of action is linear sliding, involving parallel anode-cathode pairs sliding against each other, which naturally lends itself to shear force detection. Several authors have demonstrated triboelectric generators in human interface applications [8,49,50,51]. While the technology is promising for future shear sensors, the production methods may not be suitable for scaled production (SID 3, 4) [50,52,53]. The ability to rapidly develop this technology and incorporate it within the selected textile was limited within the project timescales due to a lack of compatibility with the chosen industrial knitting machine (SID 5).

2.2.2. Capacitance

Capacitance-based sensors implemented through yarns were investigated. These can be implemented directly through knitting or as structures embroidered onto a textile surface. Similar efforts for other applications have investigated embroidery to produce pressure sensors in bedding [54] and chairs [55]. The possibility of implementing alternating rows within the knit structure itself was investigated. Fobelets et al. demonstrated the approach in their 2019 paper [56]. The requirement for compatibility with a mass-production knitting machine limited the ability to implement this method (SID 5) as it severed yarns when changing types, interrupting electrical pathway continuity. Manual reconnection of yarns would be impractical across large production volumes. Additionally, localisation (SID 6) requires the discrete distribution of sensing units across a sock, whereas the process set out by Foblets et al. provides a global response. Given the issues observed with capacitance-based approaches, a yarn-based inductance approach was excluded.

2.2.3. Resistive: Yarns

The authors could not source commercially available yarns with a resistive strain response. Within the e-textiles academic literature, custom yarns with a resistive response have been produced by layering conductive and nonconductive elements [57,58,59], surface coatings [60,61,62] and electro-spinning of PEDOT [63]. While each method demonstrates its ability to produce a resistance response under strain, they do not fit within the project specification. Namely using commercially available material without significant or expensive processing equipment (SID 3) [61]. Preliminary investigation identified that the commercially available conductive yarns would not be compatible with the required manufacturing processes (SID 5), due to needle gauge specifications [64]. Similarly, the development of a custom yarn lies outside the project’s scope (SID 3) and has been achieved by others in the literature [57,58,59,60,61,62,63]. Additive methods were sought to be implemented, including silicone-encapsulated conductive materials [65] and doping methods/techniques [66,67]. To narrow the scope of the investigation, two conductive materials were selected: carbon and silver.

2.2.4. Resistive: Carbon Particles

Carbon is well established within the sensor development community with variants such as nanotubes (CNTs) [68,69,70,71]., nano-fibres (CNFs), carbon grease [72], carbon black powder (CB)[65,72,73], inks [74] and graphene [75] well researched within the literature. Carbon grease was not investigated as wanted a solid sensor was required to mitigate the risk of leaking in the event of failure (SID 2). Although a potential solution, graphene requires high-cost equipment for its production (SID 3). Inks were not included as future work will require washing the sensor, and inks are more susceptible to washing off.

CNT have produced a repeatable compression response [68,69,70,76]. While an effective sensing response method, CNTs require extensive clean laboratory infrastructure due to their inherent toxicity. Carbon Nanotubes, therefore, fail to meet the biocompatibility requirements due to leaching if the sensor becomes damaged (SID 2). Additionally, the manufacturing environment produces scalability challenges during early development (SID 4).

CNF were implemented, achieving conductivity; however, carbon structures are known to inhibit the curing of platinum-based silicones such as Ecoflex 00-30 [Smooth-on Inc., PA, USA]. It was observed that achieving a mixing ratio suitable for conductivity resulted in cure inhibition. Therefore, the CNF-silicone paste was encased within silicone. When tested with a voltage divider setup, a voltage strain response was observed during tension-release cycles. However, due to the paste’s non-solid nature, relaxation within the structure outside the movement was observed, reducing the voltage detected (Figure 2). While the shearing of the CNF paste could produce a response, inconsistent results are likely once a normal force is applied. Importantly, with up to $1.5\times$ bodyweight being put through the foot during gait, there is a significant risk of rupture (SID 1, 2).

CB has been successfully implemented in the production of stretch sensors where the carbon black is set within silicone [65,72,73]. CB’s use in this application was investigated, and it is significantly safer and has a lower cost than CNTs. Attempts were made to recreate the work of Zhu et [65] and Muth et al. [72], who created sensors using a CB mixture embedded within a silicone substrate. CB can typically be sourced in three sizes: 20 nm, 50 nm and 75 nm. The initial investigation used 20 nm and 75 nm carbon black (CB) [SGL Carbon GmbH, Germany]. It was observed that 20 nm was not sufficiently conductive; the small particles lacked electrical continuity once mixed with silicone. Conductivity could only be achieved using high ratios of CB that inhibited silicone curing. Conversely, 75 nm produced a curable silicone but yielded high resistance (1-10 M$\mathrm{\Omega }$ range) with poor sensitivity, making it challenging to adopt as a sensor. Therefore, the 50 nm CB used by other researchers [65,74] was selected as the middle ground regarding property response. The 50 nm CB was taken forward as a potential solution. In parallel to the investigations discussed so far into yarn-based and carbon-based approaches, the possibility of a silver-based sensor was investigated.

2.2.5. Resistive: Silver Particles

Selected for its high conductivity and anti-bacterial properties, silver exhibits a higher conductivity when used as a sensing element. Three silver-based methods were investigated: inks, flakes and adhesives.

Silver conductive inks are typically used by doping a substrate [77]. Preliminary tests found the silver ink [Thermo Fisher Scientific, MA, USA] saturated the knitted base layer, providing excellent conductivity but little to no change detectable in resistance with strain. Therefore, it gives a poor ability for use as a sensor. With future development, silver inks could offer an excellent method for connecting sensors and electronics along the sock’s length. However, consideration would need to be taken regarding the washing and encapsulation of doped regions.

Another method reported within the literature involves using silver flakes suspended within a silicone substrate [78,79]. The silver flakes are typically coated in a lubricant for storage and dispensing, preventing the flakes from aggregating. The manufacturing process involves heating samples above 250°C to remove most of the lubricant [80]. Reaching similar levels for textiles such as Lycra (≈ 230–270°C) and nylon (≈ 255°C), commonly used in sock production. Matsuhisa et al. and Yoon et al. heated their samples to trigger the thermal decomposition of the lubricants at temperatures of 120 °C [78], 130°C [79] and 180°C [81]. While conductivity was achieved, it was only achieved at 250°C, which takes the temperature too close to the estimated melting point. The high mixing ratios required to achieve conductivity, related cost, and cure temperature meant that using silver flakes in their basic form fell outside the application scope (SID 4, 5).

Therefore, investigations focused on products that included silver flakes that could be cured at a lower temperature, selecting a silver adhesive [Dycotec DM-SAS-10010, Dycotec Materials Ltd., Calne, UK]. Designed to replace the solder in wearable technologies, the conductive adhesive provides the secure connection of components and the elasticity to absorb some movement that would typically cause fractures in a solder-based joint, leading to delamination or connection quality reduction. In its typical application, the changes in resistance would be minimal. A strain-dependent resistive response was identified, with the conductive adhesive selected for further investigation and evaluation of its suitability in sensor production. The conductive adhesive is supplied in a syringe, aiding manufacturing through 3D printing with a fine resolution.

2.2.6. Evaluation Summary

In Section 2.2, sensing technologies were evaluated against the specification (Table 1). Two potential candidates were identified to be taken forward: 50 nm CB and conductive adhesive. While the CB options provide an effective sensing method, it is a known carcinogen and poses risk to an already sensitive tissue boundary (SID 2). In contrast, silver has antimicrobial properties and low toxicity [82]. The conductive adhesive used in this study is commercially available and meets the higher TRL solutions (SID 3) and scalability requirements (SID 4). Thus, conductive adhesive was selected as the basis for this work. The sensor will follow the established strain gauge structure, providing sensitivity in an isolated direction of movement (Figure 3). The sensor will be manufactured through 3D printing with the supplied syringe connected to a pneumatic dispenser. A parametric design study was undertaken to identify the final design’s dimensional properties.

3. Sensor Development

The sensor development was undertaken in three stages: 1) the production of a manufacturing platform, 2) a preliminary investigation via parametric testing to identify the sensor’s form factor and 3) a secondary investigation focusing on improving sensor robustness.

3.1. Design and Manufacture

The sensor was manufactured using a 3-axis CNC linear stage with a G-code controller [WorkBee Z1+ CNC, Ooznest, Brentwood, UK], selected to provide a modular system with precise control (±0.1 mm accuracy) for repeatability in sensor production. The CNC platform was augmented with a precision syringe dispenser [Ultimus V High Precision Dispenser, Nordson, Westlake, OH, USA]. The dispenser was selected for precise control and trigger system, aiding automation and production repeatability. The CNC and dispenser were linked to enable automated dispensing via a G-code trigger mechanism, as shown in Figure 4.

While this project’s ultimate objective is to print sensors onto a knitted sock fabric, a woven lycra substrate was selected during development to speed the development process. A laser cutter [VLS3.60DT, Denford Ltd, Brighouse, UK] was used to cut uniform $20\times 90$ mm Lycra patch samples for printing. Each patch was secured using a laser-cut jig (5 mm acrylic) onto the CNC base plate. The CNC Z axis was then zeroed to the plate surface using a touch probe.

The sensors were designed in CAD [Solidworks, Dassault Systèmes, France] to facilitate parametric design adaptation. Designs were exported as DXF files and imported to a G-code generator [Cut2D, Vetric, Redditch, UK] for printing. Sensors were printed by loading pre-filled syringes of conductive adhesive into the dispenser, fitted with a 22 gauge (0.41 mm) leur dispensing tip [Adheisive Dispensing, Milton Keynes, UK]. An iterative process identified the optimum operating parameters for printing, dispensing at 4.5 PSI while the print head was moved at a constant 30 mm/min with a pass depth of 0.25 mm. Each sensor was made of parallel overlapping tracks. Sensors were cured for 30 minutes with cure temperature further investigated in Section 3.2.

3.2. Design Parametric Testing

The CNC manufacturing approach provided a repeatable basis for iterative design optimisation through parametric investigation. Three parameters were investigated to help improve sensor performance. First, the number of turns within the strain gauge style design is explored to enhance sensitivity to an applied strain. Tree, five and seven turns were selected, with an odd number chosen to produce polar endpoints for strain testing (Figure 5.a). Secondly, sensor length was investigated to understand the impact on perforamnce, with potential benefits linked to increased sensor area encompassing more potential strain. Lengths of 10, 20 and 30 mm were selected (Figure 5). The manufacturer quotes the conductive adhesive’s cure temperature as influencing stiffness [83]. The affect on performance was investigated, 80, 90 and 100°C were selected as the lower end of the recommended cure range of 80-120°C range. Greater elasticity may benefit performance; therefore, 75°C cure temperature, below the recommended levels, was selected to assess any difference. A base setup was chosen as the default case during parameterisation, from which a single variable would be altered. The sample consisted of a 30 mm long sensor with seven turns cured for 30 minutes at 80°C.

All samples were prepared using the manufacturing processes described in Section 3.1 and shown in Figure 4. Three samples were produced for each parametric variation, and five test repeats were taken for each sample (total N = 45). Preliminary testing produced dogbone samples of the conductive adhesive to undertake stress-strain testing according to BS EN ISO 527-3:2018. Due to the amount of conductive adhesive required and material cost, only two samples were tested for destruction to provide an estimate for the conductive adhesive’s elastic region. This provided the necessary data to identify appropriate test parameters, determined as 10% extension as the elastic limit for 80°C cured conductive adhesive.

Testing was undertaken on a single-axis universal load tester [Instron 5943, High Wycombe, UK] equipped with 250 N pneumatic jaws [Instron 2712-052] to safely secure samples and a 500 N load cell [Instron 2580-500N]. Quasi-static testing was undertaken between 0 and 10% of the sensor’s length at intervals of 1%. The samples were of different lengths, resulting in a different extension value. A 20-second pause was added between loading periods with a 40-second pause at the start, peak and end (Figure 6).

Data was collected using a data acquisition board [MAX31865, Adafruit, NY, USA] connected to a microcontroller [Arduino Mega, BCMI, WA, USA], which streamed data to a custom user interface [Processing IDE]. Data processing and analysis were undertaken in Matlab [v2024, MathWorks, MA, USA]. A data synchronisation signal marked the start and end of the load operation to align the independent datasets. A low-pass filter (f = 2 Hz) was applied to attenuate high-frequency noise from the data. Given the low speeds under investigation, a passband of two was selected. A moving average filter was applied using convolution to smooth the data using a window width of four. Following filtering the data was normalised against the change in resistance and starting resistance to give $\Delta R/R_0$. Following the quasi-static loading regime, an average was taken for each window defined as the 20s or 40s static regions (Figure 6).

A one-sample Kolmogorov-Smirnov test highlighted that the data was not normally distributed. The Kruskal-Wallis test was applied to determine any significant difference between data sets (Figure 7.a). There was no significant difference between the results with P values of 0.597, 0.923 and 0.827 when comparing three and five, three and seven, and five and seven turns, respectively. The response curve shows that three and five turns produce a more consistent response with standard deviations of 0.561 and 0.384, respectively. Five turns were selected with the most consistent $\delta R/R_0$ response.

Overall, the three samples returned similar results (Figure 7.b). The Kruskal-Wallis test was applied and found no significant difference between values with P = 0.761, 0.240 and 0.634 for 10 vs 20 mm, 10 vs 30 mm and 20 vs 30 mm, respectively. Given the lack of a significant difference in response between lengths, 10 mm was selected as it provides the highest sensor density potential on a given surface, leading to localisation across an array (SID 6).

The $\delta R/R_0$ response is dependent on cure temperature and the associated stiffness change. Shown with a log scale on the Y-axis (Figure 7, there is a significant difference (P = 0.028) between the 75°C and 100°C datasets. The cure temperatures are paired, with a high similarity between 75°C and 80°C (P = 0.986) and 90°C and 100°C (P = 0.970). Although they did not reach statistical significance the 75°C and 90 °C (P = 0.090), 80°C and 90 °C (P = 0.191), and 80°C and 100°C (P = 0.070) there are visible differences. From these results a curing temperature of 75°C was selected. Together, these produce the following parameters for a base sensor design (Table 2).

3.3. Robustness Improvements

After parameteric testing, the sensor produced viable performance but exhibted signification variation between samples, with coefficients of variance of 0.62±0.045. Accordingly, two approaches were investigated to improve sensor robustness: increasing the conductive adhesive quantity and reinforcement using a secondary material. Increasing conductive adhesive while improving sensor robustness significantly increases conductivity, reducing the sensor operating range.

Three methods were identified to increase conductive adhesive, double-width (DW; Figure 8.a), expanding the trace width with a second pass; double-layered (DL, Figure 8.b) where two single-width sensor layers are stacked vertically; and a combination of the two which will be referred to as the Quad setup (Figure 8.c).

The same quasi-static test method and processing (Section 3.2) was used to assess robustness improvements, using a single sample tested across five repeats for each technique (n = 15; Figure 8.d). The addition of conductive adhesive reduced the sensitivity to strain with $\delta R/R_0$ dropping from the 1-1.5 range compared to the peak of 2-3.5 observed during parametric testing. Although not significantly different (P = 0.440), the selection of DW vs DL appears to influence sensitivity at low strain levels, although the response is lost as strain increases. Interestingly, although there was no significant difference between DW and Quad (P = 0.312), a significant difference was observed between the DL and Quad samples (P = 0.020). The addition of conductive adhesive improved the coefficients of variance to 0.222 ±0.107 (DL), 0.257 ±0.133 (DW) and 0.296 ±0.111 (Quad). The Quad (Figure 9) was selected to be taken forward due to the superposition response of DW with a low strain response and DL with a high strain response. In the final sensor design, the corners were filleted to reduce the risk of crack propagation at turn points.

3.4. Lycra Performance Evaluation

Three phases of testing were used to characterise sensor performance. The first sought to investigate the influence of static drift and pressure on sensor performance. The second repeated the previously used quasi-static loading regime. Finally, sensor performance under cyclic loading conditions was investigated, the most representative of foot loading during gait.

Quasi-static testing of the final quad design highlighted no significant difference between sample responses in the Kruskal-Wallis test, with an average P value of 0.791 (Figure 10). The average percentage deviation from sample one of 11.38% demonstrates the repeatability achieved in sensor manufacturing and the improved robustness of the quad form factor.

Sensors such as force-sensitive resistors exhibit resistance drift when unloaded. The sensor response was, therefore, investigated. Three samples were left for 10 minutes without any applied load for the first test to identify a static drift, a common occurrence in resistance-based sensors. The test was repeated five times for each sample. The sensor exhibits a gradual negative drift of $0.16\pm 0.53\mathrm{\Omega }$ across the 10-minute interval, equating to 0.037% of the total range. This is due to a relaxation effect observed in the polyurethane substrate. A similar response would be seen if a silicone-based substrate was used.

In the assessment of gait and plantar loading, cyclic loading provides a representative test scenario. The sensor was evaluated over 50 cycles at 10 mm/min with three consecutive sets, representing a short walking burst. The initial investigation highlighted that cyclic strain between 0 and 10% resulted in an upward resistance drift at increasing rates (Figure 11).

3.5. Mechanism of Action

Although there is a return to baseline during unloading, when loading resumes the resistance response returns to the endpoint of the previous set (see Figure 11.b and Figure 11.c), suggesting a failure-related mechanism. The mechanism of action behind the gradual upward dynamic drift in resistance sensing response was investigated. Given that increasing the cure temperature increases conductive adhesive stiffness, it was hypothesised that this upward drift may relate to micro-fractures within the elastic polyurethane substrate of the conductive adhesive. A scanning electron microscope (SEM) was used to image lycra samples in relaxed and stretched states (Figure 12).

The SEM imaging shows a uniform surface with no evident deformation in its unstrained form (Figure 12.a and .b) even though the sample was tested before SEM and the drift response observed (Figure 11). This demonstrates the return to a baseline resistance observed when strain is removed (Figure 11). When the sample is clamped in a position of approximately 10% strain, surface changes are visible at 50× magnification (Figure 12.c). At increased magnification, there are clear strain patterns with the polyurethane substrate. More significantly, tears were observed within the substrate creating voids (Figure 12.d). These micro tears were observed down to lengths of 5.15 ±0.24 $\mu m$ with widths of 1.68 ±0.09 $\mu m$ (Figure 12.e). Similar mechanisms of action have been observed by Yoon et al. [79], who described "conductive bridges" surrounding cracks enabling the sensor to continue performing. Similarly, Shen et al. [84] observed crack formation in flexible sensor substrate above 10% strain.

3.6. Reinforcement of Sensing Element

Having identified a mechanism of action, sensor robustness improvemetns were sought . These started with reassessing the conductive adhesive’s cure temperature. The manufacturer recommendations are between 70°C and 120°C. However, with a known stiffness relationship and a goal of maximising elasticity, cure temperatures between 50°C and 80°C at 10°C intervals were investigated.

As shown in Figure 13.a and b, the lower curing temperatures improve the response, significantly reducing the drift compared to baseline. This suggests cure temperatures of 70°C and 80°C exhibit the response linked to micro-fractures (Figure 13.c and d). It is expected that the manufacturer sought stiffer connections for electronics where movement is kept minimal in comparison to the dynamic nature of this application. Although there is no significant difference between 50°C and 60°C, the lower cure temperature was selected for potential elastic capacity.

3.6.1. Knitted Substrate Implementation

To facilitate rapid prototype development and ensure a uniform substrate, a Lycra substrate was employed because it could be laser cut to produce identical samples. The approach was then translated to a knitted substrate, with sensor samples produced using the same methods as for Lycra, including dispensing rate and movement speed (Section 3). In addition, SEM was used to investigate how the conductive adhesive integrated with the different substrates, as shown in Figure 14. (Figure 14.a) highlights the smooth and uniform nature of the Lycra yarn and the low interstitial spaces between the filaments which preclude mechanical interlock forming, resulting in low conductive adhesive adhesion. Highlighted by the minimal coating found along the edge in the $1500\times$ magnification (Figure 14.c). In contrast, the knitted natural fibres have protruding fibres that provide interstices distributed across the textile surface (Figure 14.b). These stray fibres and a higher visible surface roughness result in improved integration, with fibres acting to reinforce the conductive adhesive to form a composite. Similarly, where contact is made between the natural fibres and conductive adhesive, there is a higher level of ingress along the fibres (Figure 14.d). During preliminary sample testing, the natural fibres formed a composite structure with the conductive adhesive reinforcing the knitted yarns. This was observed in the need to strain the sample to 20% to achieve the same resistive response observed within the 0-10% range in the Lycra samples. Additionally, a reduction in drift was observed, it is hypothosised that composite structure reduced the frequency of micro-tears in addition to the improved adherance observed in Figure 14.d, improving overall robustness.

3.6.2. Sensor Encapsulation in Silicone

During use, the sensor requires protection from outside elements, such as sweat, that would create an electrical short. Therefore, the cured sensor were encapsulated within silicone. In addition, the encapsulating silicone can be selected to fine-tune the force response of the strain sensor (by selecting softer or stiffer silicones). A range of biocompatible commercial silicones were selected (Exoflex 00-20, 00-30 and 00-50) for assessment in this context.

The silicones reinforcing the sensor resulted in a statistically significant difference (P<0.05) in resistance response. Being the least stiff, the 00-20 silicone provided the least support for large resistances reported. The 00-50 silicone responded within a similar range as 00-30. However the 00-30 silicone was selected due to more consistent response trajectory between samples with an intersample standard deviation of 0.139, compared to 0.299 and 3.95 for the 00-50 and 00-20, respectively.

4. Final Sensor Design and Evaluation

This section evaluates performance of the finalised sensor following the development reported in Section 3.

The finalised design incorporates six turns, adding an additional turn (with respect to earlier prototypes) to achieve level termination typical of conventional strain gauge design, as shown in Figure 16. The sensor follows the "quad" design with a length of 10 mm. The sensor was reinforced with an encapsulating layer of Ecoflex 00-30. Testing was undertaken using a cyclic loading regime across 100 cycles, which was repeated three times for each of three samples using the same setup as Section 3.2. The sample was clamped within the pneumatic jaws at a distance of 10 mm, allowing the sensor to be entirely located between the jaws. This baseline start position is equal to a 0% strain. The strain was applied to the sensor at a rate of 10 mm/min, with strain moving between 1.5% and 7.5%. The pre-strain was added to represent the slight stretch present when wearing a textile garment. Data analysis extracted sensitivity, dynamic drift, and hysteresis metrics to assess sensor performance. An example data set can be seen in Figure 17 where the left axis (blue) represents the cyclic strain loading regime and the right axis (orange) represents the resistance measured across the sensor.

Although significantly reduced following the sensor development above, some drift remains in the sensor response, evident with an increase in the peak value across the 100 cycles (Figure 17.a). For this study drift is defined as the increase in resistance per cycle relative to the total measurable range (430 $\mathrm{\Omega }$). The three samples exhibited a dynamic drift of 0.039±0.014%, 0.045±0.021% and 0.045±0.019%, respectively. Relative to the chip’s measurable range (0-428$\mathrm{\Omega }$), the sensor produces an output within the operating range of the data acquisition system for approximately 2275 (±562) cycles before the chip reaches saturation.

The final sensor, including encapsulation, operates with a reliable linear response in the 0-6% range (Figure 17). While lower than the response exhibited in preliminary testing, this is sufficient for the requirements of this application. Based on this strain limit, the sensor’s sensitivity was calculated with the linear best fit gradient of strain between 1.5% and 5%, where the response is linear (Figure 18.c).

An increase in sensitivity was observed in each of the three samples over the 100-cycle test regime (Figure 19.a); this is in keeping with the drift characteristic and the mechanism of action discussed in Section 3.5. Overall, the responses follow similar trajectories, converging at a sensitivity of approximately 0.3. However, there are significant differences between the average cyclces between sample 1 and sample 2&3 (P<0.05), with no significant difference between samples 2 and 3 (P = 0.607), highlighting the need for sensor-specific calibration.

Hysteresis was calculated using trapezoidal numerical integration [trapz, Matlab 2024b, MathWorks, MA, USA] of the response to determine the area within the hysteresis loop (Figure 19.b). The output was normalised against the upward loop to represent hysteresis as a percentage (Figure 19.c). Overall, the response follows a trajectory of a gradual increase followed by a plateau. The plateau occurs within the first 25-50 cycles, suggesting a ’bedding-in’ process. This is consistent with the related drift and sensitivity changes that manifest due to micro-fractures occurring over time.

5. Discussion

Developing strain sensors that convert mechanical strain to a measurable response is a promising solution to detecting a key causative component of DFU formation. Implementing a printed sensor using a conductive adhesive provides a practical solution for this application. A key novelty is the ability to apply the conductive adhesive repeatably to a knitted substrate. Integrating strain-sensing technology within a knitted garment, such as a sock, provides the potential for an unobtrusive monitoring method appropriate to this sensitive environment. The proof-of-concept sensor evaluation highlights commercially available conductive adhesive as a promising sensing material. While some dynamic drift was observed under cyclic loading, this is common in flexible stretch sensors. Similar performance and characteristics are reported in other textile-based sensors. For example, Li et al. [85] observed a gradual increase in $R/R_0$ over 500 cycle test regimes. Similarly, Yoon et al. [79] used sensors based on silver flakes and nano-particles encapsulated within silicone rubber and experienced dynamic drift during cyclic testing due to microfractures or "conductive bridges".

Reflecting on the specifications identified in collaboration with patient and clinical engagement (Section 3) and set out in Table 1, the developed soft sensor meets the specification criteria. The sensor does not exhibit fracture under load, with silicone reinforcement for added protection (SID 1). The silicone enclosure is safe for skin contact, and silver is naturally anti-microbial in the case of failure (SID 2). During development, commercially available materials were pioritiesed for investigation, selecting the commercially available conductive adhesive (SID 3). While the 3D printing method may not be suitable for conductive adhesive mass production, other processes, such as screen printing, could be implemented. Therefore, the conductive adhesive has sufficient scope for mass production processes with further development (SID 4). The conductive adhesive’s integration was successfully demonstrated within a knitted substrate produced on the industrial knitting machine (SID 5); future work will investigate the integration of electronics and data acquisition. Through 3D printing and parametric design choices, a sensor small enough to make an array with future development was produced, allowing for strain localisation (SID 6). Finally, with print layers of 0.25 mm, the sensor has a low profile, with the ridges produced by the strain gauge design filled with silicone to provide a continuous surface (SID 7).

It should be noted that there are some limitations due to unknown factors in the application, such as the expected strain. It was identified that the sensor can work repeatably and reliably up to 6% strain during dynamic movement. Future work will evaluate how this relates to plantar strain characteristics which are currently not well reported or understood in the literature. This sensing approach has the potential to help better understand plantar strain regimes in healthy and diabetic foot contexts, and thus advance technology for instrumentation to aid DFU prevention.

6. Conclusions

In the first half, this article presents the conceptualisation and development of a strain sensor for use in e-textiles and DFU prevention. Given the application, the limited number of appropriate sensing methodologies provided focus, guiding the research. Development highlighted good single-sensor repeatability was observed with no sigificant difference observed between repeat testing. The sensor has an estimated life-span of 2275(±562) cycles. The sensor proposed in this paper provides a promising first step towards a new wearable systems for the prevention of DFUs, meeting the specification set out in this work (Table 1). With further development this techonology has potential to significant impact patient quality of life while reducing healthcare sector expendature. Outside of healthcare the technology offers potential benefit to other fields including but not limited to general shoe fit, informing accessories shearing such as backpack straps and sports performance monitoring.